Detection of Abdominal and Mediastinal Lymph Nodes in Computed Tomography Images Using Transformer Based Deep Learning Approach

Accurate disease diagnosis and treatment planning for lymph node (LN)-related conditions depend on the precise detection of LN s in radiological modalities, such as computed tomography (CT) images. However, due to the sparse distribution of LN s, their varying sizes, low contrast, and tendency to form clusters, manual detection in CT images becomes challenging. Existing methodologies have not demonstrated sufficiently promising results to be considered as significant solutions. Hence, there is a need for a more reliable method for the automatic detection of LN s in CT images. In this paper, the authors propose a state-of-the-art transformer-based deep learning approach for detecting mediastinal lymph nodes (MLN s) and abdominal lymph nodes (ALN s) in CT scans. To demonstrate the efficacy and usefulness of the proposed methodology, the authors conducted comparative experiments with popular deep learning models such as YOLOv8 and SAM. With the proposed methodology, we achieved recall and precision rates of 98% for ALNs. For MLNs, the recall rate was 99% and the precision rate was 98%. We were able to obtain a 98.73% DS (dice score) for the mediastinal lymph nodes and a 99% DS (dice score) for the abdominal lymph nodes based on the testing images.